DP-750 environment and compute explained
Set up and configure an Azure Databricks environment is worth 15–20% of DP-750. It has two halves: choosing and configuring compute, and creating the Unity Catalog objects your data lives in. The questions are choice-based: given a workload and a constraint, which compute type, setting or object type fits. If you can say why a nightly job should not run on an all-purpose cluster, you already understand half of this domain.
Select and configure compute
The outline names five compute types. Know what each is for.
| Compute | Best for | Watch out for |
|---|---|---|
| Serverless | Notebooks, jobs and pipelines where you want no cluster management and fast start | Fewer low-level settings to tune |
| Job compute | Scheduled production jobs; created for the run and released afterwards | Not for interactive work |
| SQL warehouse | SQL queries, dashboards and BI tools | Not a general Python environment |
| Classic compute (all-purpose) | Interactive development and exploration | Charges accrue while it runs idle |
| Shared compute | Several users on one resource with isolation | Access mode limits some features |
A common exam pattern: a workload is scheduled, unattended and predictable, so job compute or serverless wins on cost. Another: many analysts run SQL concurrently, so a SQL warehouse (often serverless) wins.
Performance settings
- Node type and count: memory-optimised for wide joins and caching, compute-optimised for CPU-heavy work, and enough workers for the data volume.
- Autoscaling: a minimum and maximum worker count so the cluster grows under load and shrinks afterwards.
- Termination: auto-terminate idle interactive compute after a short period to avoid paying for it.
- Pools: keep idle ready instances to shorten start times for clusters and jobs.
Feature settings
Photon is the vectorised query engine that speeds up SQL and DataFrame operations on Delta tables, at a higher per-hour rate. Choose the Databricks runtime version deliberately: long-term support versions for stability, newer ones for new features. A machine learning runtime adds ML libraries and is not needed for ordinary ETL.
Libraries and permissions
Libraries can be installed at the compute level, so every notebook on it sees them, or scoped to a notebook. Compute permissions follow a simple ladder: attach to, restart, and manage. Granting the lowest permission that does the job is usually the answer.
Create and organise Unity Catalog objects
Unity Catalog uses a three-level namespace: catalog, schema, then table, view, volume or function. The outline asks you to create each of them from requirements.
Naming and isolation
A typical layout isolates environments with separate catalogs (dev, test, prod) or with schemas inside one catalog, and puts data meant for external sharing in its own catalog so nothing is shared by accident. When a question stresses isolation, the answer is usually a separate catalog.
Tables, views, volumes
| Object | Use for |
|---|---|
| Managed table | Data whose lifecycle Unity Catalog controls; dropping the table removes the data |
| External table | Data stored in a location you manage; dropping leaves the files |
| View | A stored query, no data of its own |
| Materialized view | A stored query result that is refreshed, for precomputed aggregations |
| Volume | Governed storage for files that are not tables, such as CSVs, images and models |
Choose external when files must outlive the table or other systems read them. Otherwise managed is the default recommendation.
DDL on managed and external tables
Expect CREATE, ALTER, DROP and COMMENT statements, and the difference in what DROP does. External tables need an external location backed by a storage credential before you can create them.
Foreign catalogs and Genie
A foreign catalog exposes an external database, such as Azure SQL Database, inside Unity Catalog so you can query it without copying data. You first create a connection that holds the host and credentials, then the catalog on top of it. AI/BI Genie instructions are text guidance you add to a Genie space, with table descriptions and example queries, so natural-language questions about your data return better answers. The outline lists this under data discovery.
Sample questions
Question 1. Forty analysts run ad hoc SQL and refresh dashboards throughout the day. The platform team wants minimal tuning and compute that scales with concurrency. Which resource should they use?
- A. A single-user all-purpose cluster running the ML runtime
- B. A serverless SQL warehouse
- C. Job compute created per query
- D. A pool of idle instances with no cluster attached
Show answer
Answer: B
A serverless SQL warehouse is built for concurrent SQL and BI workloads and manages scaling and start time for you. A single-user all-purpose cluster serves one person and idles expensively. Job compute exists for the duration of a run and cannot serve interactive dashboards. A machine learning runtime cluster adds libraries the analysts do not need.
Want more questions like this? Full DP-750 practice tests →
Question 2. Parquet files in Azure Data Lake Storage are read by several tools besides Databricks. You register them as a table in Unity Catalog, and the files must remain in storage if the table is ever dropped. What should you create?
- A. A managed table
- B. A materialized view
- C. A foreign catalog
- D. An external table on an external location
Show answer
Answer: D
An external table points to data in a location you manage, and dropping it removes only the metadata, leaving the files. A managed table deletes its data on drop. A materialized view stores computed results, not source files. A foreign catalog exposes a database and does not govern files in a lake.
Want more questions like this? Full DP-750 practice tests →
Question 3. Reference data lives in an Azure SQL database. Analysts must query it from Databricks with Unity Catalog governing access, without copying the data into the lakehouse. What should you configure?
- A. A connection and a foreign catalog
- B. A nightly copy job into a managed Delta table
- C. A volume that holds exported CSV files
- D. Auto Loader reading from the database
Show answer
Answer: A
A connection to the Azure SQL database plus a foreign catalog on top of it lets Unity Catalog govern queries against the source in place. Loading the data into Delta duplicates it, which the requirement rules out. A volume is for files, not database tables. Auto Loader ingests files from storage and does not connect to a relational database.
Want more questions like this? Full DP-750 practice tests →
What to practise
Create one catalog with a managed and an external table, drop both, and see what happens to the files. Create a serverless SQL warehouse and a job compute configuration, and compare the settings pages. Set up a foreign catalog against any database you have. Then read the Unity Catalog security guide for the next domain.